OpenAI connects Epic EHR and public health data to ChatGPT for Healthcare

Ninety-nine point one percent. That is the share of AI-generated clinical responses that physicians rated as safe across 4,363 evaluations of ChatGPT working with live electronic health record data. For a technology still earning its place in the clinic, that figure matters. And it is central to OpenAI's argument for its latest push into healthcare AI.

On September 1, 2026, OpenAI announced two new capabilities for ChatGPT for Healthcare: a direct integration with Epic, one of the world's most widely used electronic health record systems, and a Healthcare Public Data plugin that connects users to nine official sources including PubMed, ClinicalTrials.gov, DailyMed, and CMS Coverage. The announcement also included details of a physician evaluation programme spanning 60 countries, 49 languages, and 26 medical specialties, with more than 700,000 model responses reviewed to date.

Early adopters include UCSF Health, whose President and CEO Suresh Gunasekaran described the integration as having "the potential to reduce time spent synthesizing data and give clinicians more time with patients." AdventHealth's Chief AI Officer Robert Purinton framed it differently: "The value of AI starts with our people."

How does it work?

The Epic integration allows clinicians to pull authorised patient information directly into ChatGPT for Healthcare. Rather than switching between appointment notes, lab results, medication lists, and specialist documentation, a clinician can ask natural-language questions such as what has changed since the patient's last visit, or which recent results are worth reviewing before today's appointment. ChatGPT summarises the relevant information and links back to the source records in the chart.

The integration supports two modes. First, EHR context in ChatGPT, where patient data is brought into the AI workspace for review and synthesis. Second, ChatGPT embedded directly into the EHR interface, so clinicians can work with AI assistance without leaving the patient chart. Both modes are designed to sit alongside existing clinical workflows rather than replace them.

The Healthcare Public Data plugin works differently. It gives teams structured access to specific records, identifiers, and versions across nine official datasets. A pharmacy team can verify the current label and warnings for a drug via DailyMed. A research team can compare eligibility criteria across actively recruiting trials on ClinicalTrials.gov. A population health team planning a diabetes prevention programme can bring together Medicare coverage data, relevant studies from PubMed, and active trial information in a single view.

Key sources available through the plugin include:

• PubMed, for peer-reviewed medical research
• DailyMed, for current drug labelling information
• ClinicalTrials.gov, for active and recruiting trial data
• CMS Coverage, for Medicare coverage policy details
• RxNorm, for standardised drug identifiers

The full workspace also connects to enterprise systems including Microsoft SharePoint, Google Drive, Salesforce, and Slack, with existing access permissions preserved. Clinical and business teams can use ChatGPT Work to produce reports, presentations, and operational analyses. Technical teams can use Codex to build or improve software supporting care delivery.

Why does it matter?

The core problem OpenAI is addressing is fragmentation. Patient context, medical evidence, coverage data, and organisational knowledge typically sit in separate systems that do not talk to each other. Clinicians and administrators spend significant time moving between these sources manually. That is time away from patients, and it introduces opportunities for error.

Connecting these sources inside a governed, HIPAA-compliant workspace changes the equation. It is not just about speed. It is about the quality of synthesis. A clinician preparing for a complex appointment can see medication changes, outstanding referrals, recent lab trends, and relevant clinical evidence together, rather than assembling that picture from five different screens.

The safety evaluation figures also matter for institutional confidence. Across 27 clinical use cases, including pre-visit reviews, medication summaries, and handoff notes, 99.1% of responses were rated safe by physicians. For connected data source queries, more than 93% received "good" or better accuracy ratings. These are not internal benchmarks. They reflect feedback from practising physicians across dozens of countries and specialties.

The context

Across the GCC, healthcare systems are under significant pressure to improve efficiency without compromising care quality. Saudi Arabia's Vision 2030 has made digital health infrastructure a national priority, with substantial investment in AI-enabled clinical tools and integrated health platforms. The UAE has taken a similar direction, with Abu Dhabi and Dubai both advancing frameworks for responsible AI use in clinical settings.

Epic is already present across major health systems in the region, which means the new integration is not a distant prospect for GCC institutions. So the question for regional health leaders is less about whether tools like this will arrive, and more about how quickly governance frameworks can be established to deploy them safely. The OpenAI evaluation model, built around physician review rather than internal testing alone, offers a template worth examining. Getting AI into the clinic is one thing. Getting clinicians to trust it is another.